plan-with-me
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill includes robust instructions to prevent the agent from making assumptions or ignoring user intent. It defines an attack surface for indirect prompt injection by reading the local codebase during the grounding phase.
- Ingestion points: Local codebase files accessed during the grounding pass in SKILL.md.
- Boundary markers: The agent is instructed to use findings to frame questions, though no specific delimiters are defined for the codebase content.
- Capability inventory: Read-only file access, git history inspection, web search, and writing a single Markdown plan file.
- Sanitization: No explicit sanitization or escaping is mentioned for codebase content included in the plan document.
- [DATA_EXFILTRATION]: The instructions include a specific guardrail against including sensitive information such as credentials, tokens, or customer records in the generated plan document. It instructs the agent to describe the shape of data instead of copying it.
- [COMMAND_EXECUTION]: The skill restricts its capabilities to read-only operations for grounding (e.g., 'git log', file reads) and prohibits commands that modify the repository, install dependencies, or run migrations.
Audit Metadata